Bimetallic alloys have emerged as an important class of catalytic materials, spanning a wide range of shapes, sizes, and compositions. The combinatorics across this wide materials space makes predicting catalytic turnovers of individual active sites challenging. Herein, we introduce the stability of active sites as a descriptor for site-resolved reaction rates. The site stability unifies structural and compositional variations in a single descriptor. We compute this descriptor using coordination-based models trained with DFT calculations. Our approach enables instantaneous predictions of catalytic turnovers for nanostructures up to 12 nm in size. Using NO decomposition as probe reaction, we identify sites on Au-Pt nanoparticles that, because of local structure and composition, yield one-to-two orders of magnitude increase in rate compared to sites on monometallic Pt. By prescribing specific sizes, morphologies, and compositions of optimal catalytic nanoparticles, our method guides experiments towards designing bimetallic catalysts with optimal turnovers.